istance 
sed by 
in the 
mation 
mprove 
m method 
S. 
^ component 
ues in 
ation 
omponent 
9. 
| computed. 
d space 
ıl values 
ie histo- 
. grouping 
d on the 
bint in 
1e following 
of the 
se of the 
is the 
is large 
this 
DATA 
lso 
int 
  
class will be deleted and we go to ii) and finally iv) compute the inter- 
classes distances. If the interclasses distance is small for two classes, 
they will be lumped together and we go to ii). 
b) Classification with human assistance 
  
The unsupervised classification does not take into account the pre- 
cious information at the disposal of the user, for instance the ground-truth. 
The classification with human assistance allows to use this information in 
order to improve the classification. 
By use of several zones correctly recognized in the image, a number 
of classes is found,for instance water, housing, spoil heaps...The centers of 
these classes are computed and located in the 2-dimensional histogram. These‘ 
centers are considered as known and fixed for these classes. We carry out a 
new classification based on the 3rd step of the unsupervised classification, 
the initial centers are the centers of known classes and that resulting from 
the previous clustering. Only the centers of unknown classes are modified in 
this classification. 
We may apply the classification with human assistance as many times 
as we like, if each classification gives us new information about classification. 
The advantages of the compound classification "unsupervised-human 
assisted" are : 
1) only the 2-dimensional histogram is stored, this storage requires 
a limited number of core memories, and 
2) the ISODATA-type clustering in the histogram is fast. This method 
is very convenient for a repetitive use of the classification with human assis- 
tance. 
2. The use of the method for urban interpretation 
It lies both in saving computer time and enabling the possibility of 
distinguishing more clusters. In the analyzed example, it has been possible to 
distinguish 19 classes, when the number of pixels reached 66.000 for Charleroi 
and 75.000 for Brussels.